Polygon counts

I'm sorry, but you are a bit to vague. What are the average and 'extreme' amount of units and buildings going to be visible? If you will only have 10 units on screen at once, they can easily be 50k a piece on recent hardware.... if you will have 1000 on screen, you obviously need less per unit ;). Also, what is your target hardware... I've got a 6800gt, so my polygon limit may be higher than yours, but I can hit 80fps with 2 million poly's on screen :P.

Well hardware wise theres no point in making a game that will only run on top end systems, given the current market I would say I would ideally like my game to run well on a 1Ghz processor with a Geforce 2. As far as how many units I expect to have in game at once I expect i would be looking at say 30 units a team max and say 50 buldings and trees per level. Then ive got all the other things like my terrain, particle system etc going down.

I just assumed there would be some kind or normal average poly count for in game RTS models?

Again thanks for any help

Another critical thing to keep in mind is how much video RAM is required for storage. For example, we found we could run 20 cars @ 35k triangle each at high frame rates, no problem, but only vid cards with over 256 MB could hold them all.

Personally, asking others isn't the answer here. Rather the answer is to get the experience needed to answer it yourself. That means getting a program up and running as fast as possible. That provides the means to experiment which is where you get the experience from so you aren't stuck in a chicken or egg situation. Get the program up and running then test models with differant levels of detail. It is just an estimate. Spend a week experimenting to get a feel for what you can eventually do. Do a little performance tweaking on it. Nothing elborate, just get rid of the obvious mistakes. Things like maybe I should have used a vertex buffer rather than feeding the individual polygons in. Do some profiling, find out where your bottlenecks are. You don't have time to retrofit every algorithm you might use into the program, but you do have time to research what algorithms are available and their performance characteristics. It shouldn't take more than a week to get a feel for what you can do given a reasonable amount of time.

Chances are the upper limit of what is currently possible is well beyond your abilities. You can most likely do far more than what some people can do. When someone gives you a number you have no idea of your skill relative to theirs. It is nothing more than a guess. It could be that if you kept at it for the next five years you could do that or you might be able to spend a week and achieve that level of performance. You want what you can actually do three, six or however many months out. That's more than you can do today, but spend two or three weeks focused in the right direction and chances are you can estimate what you'll be able to do in three or six months pretty well.